1#include <linux/fanotify.h>
2#include <linux/fcntl.h>
3#include <linux/file.h>
4#include <linux/fs.h>
5#include <linux/anon_inodes.h>
6#include <linux/fsnotify_backend.h>
7#include <linux/init.h>
8#include <linux/mount.h>
9#include <linux/namei.h>
10#include <linux/poll.h>
11#include <linux/security.h>
12#include <linux/syscalls.h>
13#include <linux/slab.h>
14#include <linux/types.h>
15#include <linux/uaccess.h>
16#include <linux/compat.h>
17
18#include <asm/ioctls.h>
19
20#include "../../mount.h"
21#include "../fdinfo.h"
22#include "fanotify.h"
23
24#define FANOTIFY_DEFAULT_MAX_EVENTS	16384
25#define FANOTIFY_DEFAULT_MAX_MARKS	8192
26#define FANOTIFY_DEFAULT_MAX_LISTENERS	128
27
28/*
29 * All flags that may be specified in parameter event_f_flags of fanotify_init.
30 *
31 * Internal and external open flags are stored together in field f_flags of
32 * struct file. Only external open flags shall be allowed in event_f_flags.
33 * Internal flags like FMODE_NONOTIFY, FMODE_EXEC, FMODE_NOCMTIME shall be
34 * excluded.
35 */
36#define	FANOTIFY_INIT_ALL_EVENT_F_BITS				( \
37		O_ACCMODE	| O_APPEND	| O_NONBLOCK	| \
38		__O_SYNC	| O_DSYNC	| O_CLOEXEC     | \
39		O_LARGEFILE	| O_NOATIME	)
40
41extern const struct fsnotify_ops fanotify_fsnotify_ops;
42
43static struct kmem_cache *fanotify_mark_cache __read_mostly;
44struct kmem_cache *fanotify_event_cachep __read_mostly;
45struct kmem_cache *fanotify_perm_event_cachep __read_mostly;
46
47/*
48 * Get an fsnotify notification event if one exists and is small
49 * enough to fit in "count". Return an error pointer if the count
50 * is not large enough.
51 *
52 * Called with the group->notification_mutex held.
53 */
54static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
55					    size_t count)
56{
57	BUG_ON(!mutex_is_locked(&group->notification_mutex));
58
59	pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
60
61	if (fsnotify_notify_queue_is_empty(group))
62		return NULL;
63
64	if (FAN_EVENT_METADATA_LEN > count)
65		return ERR_PTR(-EINVAL);
66
67	/* held the notification_mutex the whole time, so this is the
68	 * same event we peeked above */
69	return fsnotify_remove_first_event(group);
70}
71
72static int create_fd(struct fsnotify_group *group,
73		     struct fanotify_event_info *event,
74		     struct file **file)
75{
76	int client_fd;
77	struct file *new_file;
78
79	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
80
81	client_fd = get_unused_fd_flags(group->fanotify_data.f_flags);
82	if (client_fd < 0)
83		return client_fd;
84
85	/*
86	 * we need a new file handle for the userspace program so it can read even if it was
87	 * originally opened O_WRONLY.
88	 */
89	/* it's possible this event was an overflow event.  in that case dentry and mnt
90	 * are NULL;  That's fine, just don't call dentry open */
91	if (event->path.dentry && event->path.mnt)
92		new_file = dentry_open(&event->path,
93				       group->fanotify_data.f_flags | FMODE_NONOTIFY,
94				       current_cred());
95	else
96		new_file = ERR_PTR(-EOVERFLOW);
97	if (IS_ERR(new_file)) {
98		/*
99		 * we still send an event even if we can't open the file.  this
100		 * can happen when say tasks are gone and we try to open their
101		 * /proc files or we try to open a WRONLY file like in sysfs
102		 * we just send the errno to userspace since there isn't much
103		 * else we can do.
104		 */
105		put_unused_fd(client_fd);
106		client_fd = PTR_ERR(new_file);
107	} else {
108		*file = new_file;
109	}
110
111	return client_fd;
112}
113
114static int fill_event_metadata(struct fsnotify_group *group,
115			       struct fanotify_event_metadata *metadata,
116			       struct fsnotify_event *fsn_event,
117			       struct file **file)
118{
119	int ret = 0;
120	struct fanotify_event_info *event;
121
122	pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
123		 group, metadata, fsn_event);
124
125	*file = NULL;
126	event = container_of(fsn_event, struct fanotify_event_info, fse);
127	metadata->event_len = FAN_EVENT_METADATA_LEN;
128	metadata->metadata_len = FAN_EVENT_METADATA_LEN;
129	metadata->vers = FANOTIFY_METADATA_VERSION;
130	metadata->reserved = 0;
131	metadata->mask = fsn_event->mask & FAN_ALL_OUTGOING_EVENTS;
132	metadata->pid = pid_vnr(event->tgid);
133	if (unlikely(fsn_event->mask & FAN_Q_OVERFLOW))
134		metadata->fd = FAN_NOFD;
135	else {
136		metadata->fd = create_fd(group, event, file);
137		if (metadata->fd < 0)
138			ret = metadata->fd;
139	}
140
141	return ret;
142}
143
144#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
145static struct fanotify_perm_event_info *dequeue_event(
146				struct fsnotify_group *group, int fd)
147{
148	struct fanotify_perm_event_info *event, *return_e = NULL;
149
150	spin_lock(&group->fanotify_data.access_lock);
151	list_for_each_entry(event, &group->fanotify_data.access_list,
152			    fae.fse.list) {
153		if (event->fd != fd)
154			continue;
155
156		list_del_init(&event->fae.fse.list);
157		return_e = event;
158		break;
159	}
160	spin_unlock(&group->fanotify_data.access_lock);
161
162	pr_debug("%s: found return_re=%p\n", __func__, return_e);
163
164	return return_e;
165}
166
167static int process_access_response(struct fsnotify_group *group,
168				   struct fanotify_response *response_struct)
169{
170	struct fanotify_perm_event_info *event;
171	int fd = response_struct->fd;
172	int response = response_struct->response;
173
174	pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
175		 fd, response);
176	/*
177	 * make sure the response is valid, if invalid we do nothing and either
178	 * userspace can send a valid response or we will clean it up after the
179	 * timeout
180	 */
181	switch (response) {
182	case FAN_ALLOW:
183	case FAN_DENY:
184		break;
185	default:
186		return -EINVAL;
187	}
188
189	if (fd < 0)
190		return -EINVAL;
191
192	event = dequeue_event(group, fd);
193	if (!event)
194		return -ENOENT;
195
196	event->response = response;
197	wake_up(&group->fanotify_data.access_waitq);
198
199	return 0;
200}
201#endif
202
203static ssize_t copy_event_to_user(struct fsnotify_group *group,
204				  struct fsnotify_event *event,
205				  char __user *buf)
206{
207	struct fanotify_event_metadata fanotify_event_metadata;
208	struct file *f;
209	int fd, ret;
210
211	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
212
213	ret = fill_event_metadata(group, &fanotify_event_metadata, event, &f);
214	if (ret < 0)
215		return ret;
216
217	fd = fanotify_event_metadata.fd;
218	ret = -EFAULT;
219	if (copy_to_user(buf, &fanotify_event_metadata,
220			 fanotify_event_metadata.event_len))
221		goto out_close_fd;
222
223#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
224	if (event->mask & FAN_ALL_PERM_EVENTS)
225		FANOTIFY_PE(event)->fd = fd;
226#endif
227
228	if (fd != FAN_NOFD)
229		fd_install(fd, f);
230	return fanotify_event_metadata.event_len;
231
232out_close_fd:
233	if (fd != FAN_NOFD) {
234		put_unused_fd(fd);
235		fput(f);
236	}
237	return ret;
238}
239
240/* intofiy userspace file descriptor functions */
241static unsigned int fanotify_poll(struct file *file, poll_table *wait)
242{
243	struct fsnotify_group *group = file->private_data;
244	int ret = 0;
245
246	poll_wait(file, &group->notification_waitq, wait);
247	mutex_lock(&group->notification_mutex);
248	if (!fsnotify_notify_queue_is_empty(group))
249		ret = POLLIN | POLLRDNORM;
250	mutex_unlock(&group->notification_mutex);
251
252	return ret;
253}
254
255static ssize_t fanotify_read(struct file *file, char __user *buf,
256			     size_t count, loff_t *pos)
257{
258	struct fsnotify_group *group;
259	struct fsnotify_event *kevent;
260	char __user *start;
261	int ret;
262	DEFINE_WAIT_FUNC(wait, woken_wake_function);
263
264	start = buf;
265	group = file->private_data;
266
267	pr_debug("%s: group=%p\n", __func__, group);
268
269	add_wait_queue(&group->notification_waitq, &wait);
270	while (1) {
271		mutex_lock(&group->notification_mutex);
272		kevent = get_one_event(group, count);
273		mutex_unlock(&group->notification_mutex);
274
275		if (IS_ERR(kevent)) {
276			ret = PTR_ERR(kevent);
277			break;
278		}
279
280		if (!kevent) {
281			ret = -EAGAIN;
282			if (file->f_flags & O_NONBLOCK)
283				break;
284
285			ret = -ERESTARTSYS;
286			if (signal_pending(current))
287				break;
288
289			if (start != buf)
290				break;
291
292			wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
293			continue;
294		}
295
296		ret = copy_event_to_user(group, kevent, buf);
297		/*
298		 * Permission events get queued to wait for response.  Other
299		 * events can be destroyed now.
300		 */
301		if (!(kevent->mask & FAN_ALL_PERM_EVENTS)) {
302			fsnotify_destroy_event(group, kevent);
303			if (ret < 0)
304				break;
305		} else {
306#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
307			if (ret < 0) {
308				FANOTIFY_PE(kevent)->response = FAN_DENY;
309				wake_up(&group->fanotify_data.access_waitq);
310				break;
311			}
312			spin_lock(&group->fanotify_data.access_lock);
313			list_add_tail(&kevent->list,
314				      &group->fanotify_data.access_list);
315			spin_unlock(&group->fanotify_data.access_lock);
316#endif
317		}
318		buf += ret;
319		count -= ret;
320	}
321	remove_wait_queue(&group->notification_waitq, &wait);
322
323	if (start != buf && ret != -EFAULT)
324		ret = buf - start;
325	return ret;
326}
327
328static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
329{
330#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
331	struct fanotify_response response = { .fd = -1, .response = -1 };
332	struct fsnotify_group *group;
333	int ret;
334
335	group = file->private_data;
336
337	if (count > sizeof(response))
338		count = sizeof(response);
339
340	pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
341
342	if (copy_from_user(&response, buf, count))
343		return -EFAULT;
344
345	ret = process_access_response(group, &response);
346	if (ret < 0)
347		count = ret;
348
349	return count;
350#else
351	return -EINVAL;
352#endif
353}
354
355static int fanotify_release(struct inode *ignored, struct file *file)
356{
357	struct fsnotify_group *group = file->private_data;
358
359#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
360	struct fanotify_perm_event_info *event, *next;
361
362	/*
363	 * There may be still new events arriving in the notification queue
364	 * but since userspace cannot use fanotify fd anymore, no event can
365	 * enter or leave access_list by now.
366	 */
367	spin_lock(&group->fanotify_data.access_lock);
368
369	atomic_inc(&group->fanotify_data.bypass_perm);
370
371	list_for_each_entry_safe(event, next, &group->fanotify_data.access_list,
372				 fae.fse.list) {
373		pr_debug("%s: found group=%p event=%p\n", __func__, group,
374			 event);
375
376		list_del_init(&event->fae.fse.list);
377		event->response = FAN_ALLOW;
378	}
379	spin_unlock(&group->fanotify_data.access_lock);
380
381	/*
382	 * Since bypass_perm is set, newly queued events will not wait for
383	 * access response. Wake up the already sleeping ones now.
384	 * synchronize_srcu() in fsnotify_destroy_group() will wait for all
385	 * processes sleeping in fanotify_handle_event() waiting for access
386	 * response and thus also for all permission events to be freed.
387	 */
388	wake_up(&group->fanotify_data.access_waitq);
389#endif
390
391	/* matches the fanotify_init->fsnotify_alloc_group */
392	fsnotify_destroy_group(group);
393
394	return 0;
395}
396
397static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
398{
399	struct fsnotify_group *group;
400	struct fsnotify_event *fsn_event;
401	void __user *p;
402	int ret = -ENOTTY;
403	size_t send_len = 0;
404
405	group = file->private_data;
406
407	p = (void __user *) arg;
408
409	switch (cmd) {
410	case FIONREAD:
411		mutex_lock(&group->notification_mutex);
412		list_for_each_entry(fsn_event, &group->notification_list, list)
413			send_len += FAN_EVENT_METADATA_LEN;
414		mutex_unlock(&group->notification_mutex);
415		ret = put_user(send_len, (int __user *) p);
416		break;
417	}
418
419	return ret;
420}
421
422static const struct file_operations fanotify_fops = {
423	.show_fdinfo	= fanotify_show_fdinfo,
424	.poll		= fanotify_poll,
425	.read		= fanotify_read,
426	.write		= fanotify_write,
427	.fasync		= NULL,
428	.release	= fanotify_release,
429	.unlocked_ioctl	= fanotify_ioctl,
430	.compat_ioctl	= fanotify_ioctl,
431	.llseek		= noop_llseek,
432};
433
434static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
435{
436	kmem_cache_free(fanotify_mark_cache, fsn_mark);
437}
438
439static int fanotify_find_path(int dfd, const char __user *filename,
440			      struct path *path, unsigned int flags)
441{
442	int ret;
443
444	pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
445		 dfd, filename, flags);
446
447	if (filename == NULL) {
448		struct fd f = fdget(dfd);
449
450		ret = -EBADF;
451		if (!f.file)
452			goto out;
453
454		ret = -ENOTDIR;
455		if ((flags & FAN_MARK_ONLYDIR) &&
456		    !(S_ISDIR(file_inode(f.file)->i_mode))) {
457			fdput(f);
458			goto out;
459		}
460
461		*path = f.file->f_path;
462		path_get(path);
463		fdput(f);
464	} else {
465		unsigned int lookup_flags = 0;
466
467		if (!(flags & FAN_MARK_DONT_FOLLOW))
468			lookup_flags |= LOOKUP_FOLLOW;
469		if (flags & FAN_MARK_ONLYDIR)
470			lookup_flags |= LOOKUP_DIRECTORY;
471
472		ret = user_path_at(dfd, filename, lookup_flags, path);
473		if (ret)
474			goto out;
475	}
476
477	/* you can only watch an inode if you have read permissions on it */
478	ret = inode_permission(path->dentry->d_inode, MAY_READ);
479	if (ret)
480		path_put(path);
481out:
482	return ret;
483}
484
485static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
486					    __u32 mask,
487					    unsigned int flags,
488					    int *destroy)
489{
490	__u32 oldmask = 0;
491
492	spin_lock(&fsn_mark->lock);
493	if (!(flags & FAN_MARK_IGNORED_MASK)) {
494		__u32 tmask = fsn_mark->mask & ~mask;
495
496		if (flags & FAN_MARK_ONDIR)
497			tmask &= ~FAN_ONDIR;
498
499		oldmask = fsn_mark->mask;
500		fsnotify_set_mark_mask_locked(fsn_mark, tmask);
501	} else {
502		__u32 tmask = fsn_mark->ignored_mask & ~mask;
503		if (flags & FAN_MARK_ONDIR)
504			tmask &= ~FAN_ONDIR;
505
506		fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
507	}
508	*destroy = !(fsn_mark->mask | fsn_mark->ignored_mask);
509	spin_unlock(&fsn_mark->lock);
510
511	return mask & oldmask;
512}
513
514static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
515					 struct vfsmount *mnt, __u32 mask,
516					 unsigned int flags)
517{
518	struct fsnotify_mark *fsn_mark = NULL;
519	__u32 removed;
520	int destroy_mark;
521
522	mutex_lock(&group->mark_mutex);
523	fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
524	if (!fsn_mark) {
525		mutex_unlock(&group->mark_mutex);
526		return -ENOENT;
527	}
528
529	removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
530						 &destroy_mark);
531	if (destroy_mark)
532		fsnotify_destroy_mark_locked(fsn_mark, group);
533	mutex_unlock(&group->mark_mutex);
534
535	fsnotify_put_mark(fsn_mark);
536	if (removed & real_mount(mnt)->mnt_fsnotify_mask)
537		fsnotify_recalc_vfsmount_mask(mnt);
538
539	return 0;
540}
541
542static int fanotify_remove_inode_mark(struct fsnotify_group *group,
543				      struct inode *inode, __u32 mask,
544				      unsigned int flags)
545{
546	struct fsnotify_mark *fsn_mark = NULL;
547	__u32 removed;
548	int destroy_mark;
549
550	mutex_lock(&group->mark_mutex);
551	fsn_mark = fsnotify_find_inode_mark(group, inode);
552	if (!fsn_mark) {
553		mutex_unlock(&group->mark_mutex);
554		return -ENOENT;
555	}
556
557	removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
558						 &destroy_mark);
559	if (destroy_mark)
560		fsnotify_destroy_mark_locked(fsn_mark, group);
561	mutex_unlock(&group->mark_mutex);
562
563	/* matches the fsnotify_find_inode_mark() */
564	fsnotify_put_mark(fsn_mark);
565	if (removed & inode->i_fsnotify_mask)
566		fsnotify_recalc_inode_mask(inode);
567
568	return 0;
569}
570
571static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
572				       __u32 mask,
573				       unsigned int flags)
574{
575	__u32 oldmask = -1;
576
577	spin_lock(&fsn_mark->lock);
578	if (!(flags & FAN_MARK_IGNORED_MASK)) {
579		__u32 tmask = fsn_mark->mask | mask;
580
581		if (flags & FAN_MARK_ONDIR)
582			tmask |= FAN_ONDIR;
583
584		oldmask = fsn_mark->mask;
585		fsnotify_set_mark_mask_locked(fsn_mark, tmask);
586	} else {
587		__u32 tmask = fsn_mark->ignored_mask | mask;
588		if (flags & FAN_MARK_ONDIR)
589			tmask |= FAN_ONDIR;
590
591		fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
592		if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
593			fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
594	}
595	spin_unlock(&fsn_mark->lock);
596
597	return mask & ~oldmask;
598}
599
600static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
601						   struct inode *inode,
602						   struct vfsmount *mnt)
603{
604	struct fsnotify_mark *mark;
605	int ret;
606
607	if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
608		return ERR_PTR(-ENOSPC);
609
610	mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
611	if (!mark)
612		return ERR_PTR(-ENOMEM);
613
614	fsnotify_init_mark(mark, fanotify_free_mark);
615	ret = fsnotify_add_mark_locked(mark, group, inode, mnt, 0);
616	if (ret) {
617		fsnotify_put_mark(mark);
618		return ERR_PTR(ret);
619	}
620
621	return mark;
622}
623
624
625static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
626				      struct vfsmount *mnt, __u32 mask,
627				      unsigned int flags)
628{
629	struct fsnotify_mark *fsn_mark;
630	__u32 added;
631
632	mutex_lock(&group->mark_mutex);
633	fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
634	if (!fsn_mark) {
635		fsn_mark = fanotify_add_new_mark(group, NULL, mnt);
636		if (IS_ERR(fsn_mark)) {
637			mutex_unlock(&group->mark_mutex);
638			return PTR_ERR(fsn_mark);
639		}
640	}
641	added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
642	mutex_unlock(&group->mark_mutex);
643
644	if (added & ~real_mount(mnt)->mnt_fsnotify_mask)
645		fsnotify_recalc_vfsmount_mask(mnt);
646
647	fsnotify_put_mark(fsn_mark);
648	return 0;
649}
650
651static int fanotify_add_inode_mark(struct fsnotify_group *group,
652				   struct inode *inode, __u32 mask,
653				   unsigned int flags)
654{
655	struct fsnotify_mark *fsn_mark;
656	__u32 added;
657
658	pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
659
660	/*
661	 * If some other task has this inode open for write we should not add
662	 * an ignored mark, unless that ignored mark is supposed to survive
663	 * modification changes anyway.
664	 */
665	if ((flags & FAN_MARK_IGNORED_MASK) &&
666	    !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
667	    (atomic_read(&inode->i_writecount) > 0))
668		return 0;
669
670	mutex_lock(&group->mark_mutex);
671	fsn_mark = fsnotify_find_inode_mark(group, inode);
672	if (!fsn_mark) {
673		fsn_mark = fanotify_add_new_mark(group, inode, NULL);
674		if (IS_ERR(fsn_mark)) {
675			mutex_unlock(&group->mark_mutex);
676			return PTR_ERR(fsn_mark);
677		}
678	}
679	added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
680	mutex_unlock(&group->mark_mutex);
681
682	if (added & ~inode->i_fsnotify_mask)
683		fsnotify_recalc_inode_mask(inode);
684
685	fsnotify_put_mark(fsn_mark);
686	return 0;
687}
688
689/* fanotify syscalls */
690SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
691{
692	struct fsnotify_group *group;
693	int f_flags, fd;
694	struct user_struct *user;
695	struct fanotify_event_info *oevent;
696
697	pr_debug("%s: flags=%d event_f_flags=%d\n",
698		__func__, flags, event_f_flags);
699
700	if (!capable(CAP_SYS_ADMIN))
701		return -EPERM;
702
703	if (flags & ~FAN_ALL_INIT_FLAGS)
704		return -EINVAL;
705
706	if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS)
707		return -EINVAL;
708
709	switch (event_f_flags & O_ACCMODE) {
710	case O_RDONLY:
711	case O_RDWR:
712	case O_WRONLY:
713		break;
714	default:
715		return -EINVAL;
716	}
717
718	user = get_current_user();
719	if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
720		free_uid(user);
721		return -EMFILE;
722	}
723
724	f_flags = O_RDWR | FMODE_NONOTIFY;
725	if (flags & FAN_CLOEXEC)
726		f_flags |= O_CLOEXEC;
727	if (flags & FAN_NONBLOCK)
728		f_flags |= O_NONBLOCK;
729
730	/* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
731	group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
732	if (IS_ERR(group)) {
733		free_uid(user);
734		return PTR_ERR(group);
735	}
736
737	group->fanotify_data.user = user;
738	atomic_inc(&user->fanotify_listeners);
739
740	oevent = fanotify_alloc_event(NULL, FS_Q_OVERFLOW, NULL);
741	if (unlikely(!oevent)) {
742		fd = -ENOMEM;
743		goto out_destroy_group;
744	}
745	group->overflow_event = &oevent->fse;
746
747	if (force_o_largefile())
748		event_f_flags |= O_LARGEFILE;
749	group->fanotify_data.f_flags = event_f_flags;
750#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
751	spin_lock_init(&group->fanotify_data.access_lock);
752	init_waitqueue_head(&group->fanotify_data.access_waitq);
753	INIT_LIST_HEAD(&group->fanotify_data.access_list);
754	atomic_set(&group->fanotify_data.bypass_perm, 0);
755#endif
756	switch (flags & FAN_ALL_CLASS_BITS) {
757	case FAN_CLASS_NOTIF:
758		group->priority = FS_PRIO_0;
759		break;
760	case FAN_CLASS_CONTENT:
761		group->priority = FS_PRIO_1;
762		break;
763	case FAN_CLASS_PRE_CONTENT:
764		group->priority = FS_PRIO_2;
765		break;
766	default:
767		fd = -EINVAL;
768		goto out_destroy_group;
769	}
770
771	if (flags & FAN_UNLIMITED_QUEUE) {
772		fd = -EPERM;
773		if (!capable(CAP_SYS_ADMIN))
774			goto out_destroy_group;
775		group->max_events = UINT_MAX;
776	} else {
777		group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
778	}
779
780	if (flags & FAN_UNLIMITED_MARKS) {
781		fd = -EPERM;
782		if (!capable(CAP_SYS_ADMIN))
783			goto out_destroy_group;
784		group->fanotify_data.max_marks = UINT_MAX;
785	} else {
786		group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
787	}
788
789	fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
790	if (fd < 0)
791		goto out_destroy_group;
792
793	return fd;
794
795out_destroy_group:
796	fsnotify_destroy_group(group);
797	return fd;
798}
799
800SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
801			      __u64, mask, int, dfd,
802			      const char  __user *, pathname)
803{
804	struct inode *inode = NULL;
805	struct vfsmount *mnt = NULL;
806	struct fsnotify_group *group;
807	struct fd f;
808	struct path path;
809	int ret;
810
811	pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
812		 __func__, fanotify_fd, flags, dfd, pathname, mask);
813
814	/* we only use the lower 32 bits as of right now. */
815	if (mask & ((__u64)0xffffffff << 32))
816		return -EINVAL;
817
818	if (flags & ~FAN_ALL_MARK_FLAGS)
819		return -EINVAL;
820	switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
821	case FAN_MARK_ADD:		/* fallthrough */
822	case FAN_MARK_REMOVE:
823		if (!mask)
824			return -EINVAL;
825		break;
826	case FAN_MARK_FLUSH:
827		if (flags & ~(FAN_MARK_MOUNT | FAN_MARK_FLUSH))
828			return -EINVAL;
829		break;
830	default:
831		return -EINVAL;
832	}
833
834	if (mask & FAN_ONDIR) {
835		flags |= FAN_MARK_ONDIR;
836		mask &= ~FAN_ONDIR;
837	}
838
839#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
840	if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
841#else
842	if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
843#endif
844		return -EINVAL;
845
846	f = fdget(fanotify_fd);
847	if (unlikely(!f.file))
848		return -EBADF;
849
850	/* verify that this is indeed an fanotify instance */
851	ret = -EINVAL;
852	if (unlikely(f.file->f_op != &fanotify_fops))
853		goto fput_and_out;
854	group = f.file->private_data;
855
856	/*
857	 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
858	 * allowed to set permissions events.
859	 */
860	ret = -EINVAL;
861	if (mask & FAN_ALL_PERM_EVENTS &&
862	    group->priority == FS_PRIO_0)
863		goto fput_and_out;
864
865	if (flags & FAN_MARK_FLUSH) {
866		ret = 0;
867		if (flags & FAN_MARK_MOUNT)
868			fsnotify_clear_vfsmount_marks_by_group(group);
869		else
870			fsnotify_clear_inode_marks_by_group(group);
871		goto fput_and_out;
872	}
873
874	ret = fanotify_find_path(dfd, pathname, &path, flags);
875	if (ret)
876		goto fput_and_out;
877
878	/* inode held in place by reference to path; group by fget on fd */
879	if (!(flags & FAN_MARK_MOUNT))
880		inode = path.dentry->d_inode;
881	else
882		mnt = path.mnt;
883
884	/* create/update an inode mark */
885	switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE)) {
886	case FAN_MARK_ADD:
887		if (flags & FAN_MARK_MOUNT)
888			ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
889		else
890			ret = fanotify_add_inode_mark(group, inode, mask, flags);
891		break;
892	case FAN_MARK_REMOVE:
893		if (flags & FAN_MARK_MOUNT)
894			ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
895		else
896			ret = fanotify_remove_inode_mark(group, inode, mask, flags);
897		break;
898	default:
899		ret = -EINVAL;
900	}
901
902	path_put(&path);
903fput_and_out:
904	fdput(f);
905	return ret;
906}
907
908#ifdef CONFIG_COMPAT
909COMPAT_SYSCALL_DEFINE6(fanotify_mark,
910				int, fanotify_fd, unsigned int, flags,
911				__u32, mask0, __u32, mask1, int, dfd,
912				const char  __user *, pathname)
913{
914	return sys_fanotify_mark(fanotify_fd, flags,
915#ifdef __BIG_ENDIAN
916				((__u64)mask0 << 32) | mask1,
917#else
918				((__u64)mask1 << 32) | mask0,
919#endif
920				 dfd, pathname);
921}
922#endif
923
924/*
925 * fanotify_user_setup - Our initialization function.  Note that we cannot return
926 * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
927 * must result in panic().
928 */
929static int __init fanotify_user_setup(void)
930{
931	fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
932	fanotify_event_cachep = KMEM_CACHE(fanotify_event_info, SLAB_PANIC);
933#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
934	fanotify_perm_event_cachep = KMEM_CACHE(fanotify_perm_event_info,
935						SLAB_PANIC);
936#endif
937
938	return 0;
939}
940device_initcall(fanotify_user_setup);
941